RELIABILITY AND ERROR RECOVERY

CHAPTER SUMMARY Reliable systems expect failure and plan for it. Retry only temporary errors, wait longer each time, add jitter, cap attempts, and make repeated actions safe. Detect loops through repetition checks, progress checks, and hard limits enforced by the harness. Handle timeouts and stale data by recording them, rechecking freshness, and being honest about uncertainty. Degrade gracefully with a fallback chain and always disclose it. Match the amount of protection to the stakes. KEY TERMS REVIEW Retry: attempting an action again. Backoff: waiting longer between attempts. Exponential backoff: doubling the wait each time. Jitter: random variation in wait times. Idempotent: having the same effect if done once or many times. Infinite loop: a repetition that never ends. Stale data: outdated information. Graceful degradation: reducing function rather than failing completely. Fallback chain: ordered backup options.